molecule-core/canvas/src/components/ProviderModelSelector.tsx
Molecule AI Core-FE 85b3e42c01
Some checks failed
CI / Canvas Deploy Reminder (push) Blocked by required conditions
Block internal-flavored paths / Block forbidden paths (push) Successful in 23s
CI / Detect changes (push) Successful in 1m12s
Harness Replays / detect-changes (push) Failing after 23s
Harness Replays / Harness Replays (push) Has been skipped
E2E Staging Canvas (Playwright) / detect-changes (push) Successful in 1m15s
E2E API Smoke Test / detect-changes (push) Successful in 1m17s
publish-workspace-server-image / build-and-push (push) Failing after 20s
Handlers Postgres Integration / detect-changes (push) Successful in 1m13s
Secret scan / Scan diff for credential-shaped strings (push) Successful in 46s
Runtime PR-Built Compatibility / detect-changes (push) Successful in 53s
publish-canvas-image / Build & push canvas image (push) Failing after 1m47s
Continuous synthetic E2E (staging) / Synthetic E2E against staging (push) Failing after 19s
CI / Platform (Go) (push) Successful in 10s
CI / Shellcheck (E2E scripts) (push) Successful in 10s
CI / Python Lint & Test (push) Successful in 12s
E2E API Smoke Test / E2E API Smoke Test (push) Successful in 15s
Handlers Postgres Integration / Handlers Postgres Integration (push) Successful in 15s
Runtime PR-Built Compatibility / PR-built wheel + import smoke (push) Successful in 8s
CI / Canvas (Next.js) (push) Has been cancelled
E2E Staging Canvas (Playwright) / Canvas tabs E2E (push) Has been cancelled
Sweep stale e2e-* orgs (staging) / Sweep e2e orgs (push) Successful in 10s
Sweep stale AWS Secrets Manager secrets / Sweep AWS Secrets Manager (push) Failing after 17s
ci-required-drift / drift (push) Failing after 10m3s
Canary — staging SaaS smoke (every 30 min) / Canary smoke (push) Failing after 5m46s
fix(canvas/test): resolve ~80 test failures across 17 test files (#299)
[core-lead-agent] lead-merge after CI green + SOP-6 tier review
Co-authored-by: Molecule AI Core-FE <core-fe@agents.moleculesai.app>
Co-committed-by: Molecule AI Core-FE <core-fe@agents.moleculesai.app>
2026-05-11 08:14:55 +00:00

524 lines
20 KiB
TypeScript

"use client";
/**
* ProviderModelSelector — single source of truth for the provider→model
* dropdown chain shared across:
* 1. MissingKeysModal (template deploy / first-time onboarding modal)
* 2. ConfigTab (per-workspace settings — Runtime section)
* 3. TemplatePalette (template side panel — inherits via MissingKeysModal)
*
* The user picks Provider FIRST (Anthropic API, Claude Code subscription,
* MiniMax, Z.ai GLM, ...). The model dropdown then filters to only that
* provider's models. Wildcard providers (huggingface/*, openrouter/*,
* custom/*) reveal a free-text model input with a tooltip explaining the
* wildcard.
*
* Provider taxonomy:
* - Multiple models can share the same `required_env` (e.g. all
* ANTHROPIC_AUTH_TOKEN-routed third-party providers — MiniMax, GLM,
* Kimi, DeepSeek). Grouping ONLY by env-tuple collapses them all into
* one bucket. We split further by vendor inferred from the model id
* so the user sees "MiniMax" and "Z.ai (GLM)" as separate options.
* - Vendor is inferred via prefix rules below. Templates that ship
* explicit vendor metadata (future) should override the heuristic.
*/
import { useId, useMemo } from "react";
export interface SelectorModel {
id: string;
name?: string;
required_env?: string[];
}
/** A provider option in the dropdown — one row corresponds to one
* vendor + env-tuple combo, holding the models that map to it. */
export interface ProviderEntry {
/** Stable id used as the <option value>. `${vendor}|${sortedEnv}`. */
id: string;
/** Inferred vendor key (e.g. "minimax", "anthropic-oauth"). */
vendor: string;
/** Human label shown in the dropdown. */
label: string;
/** Env vars required by every model in this provider. */
envVars: string[];
/** Models bucketed under this provider. */
models: SelectorModel[];
/** True when ANY model id contains "*" — UI shows free-text model input. */
wildcard: boolean;
/** Optional tooltip text (rendered as native title=). */
tooltip?: string;
}
export interface SelectorValue {
/** ProviderEntry.id of the selected provider. Empty string = nothing
* picked yet (parent should treat as invalid for save). */
providerId: string;
/** Selected model slug. For wildcard providers this is whatever the
* user typed in the free-text input. */
model: string;
/** Snapshot of envVars from the selected provider. Re-emitted on every
* change so consumers can re-render credential fields without
* re-inferring from the model. */
envVars: string[];
}
interface Props {
models: SelectorModel[];
value: SelectorValue;
onChange: (next: SelectorValue) => void;
/** Display variant. "grid" = label+control side-by-side (used in ConfigTab
* Runtime section). "stack" = vertical (used in MissingKeysModal). */
variant?: "grid" | "stack";
/** When true, parent caller is opting in to power-user free-text. Adds a
* "Custom (type model id)..." escape-hatch entry as a model option even
* when the chosen provider isn't wildcard. ConfigTab uses this; the
* deploy modal does not. */
allowCustomModelEscape?: boolean;
disabled?: boolean;
/** Optional id-prefix for label↔control wiring (WCAG 1.3.1). Default
* uses useId(). */
idPrefix?: string;
}
// -----------------------------------------------------------------------------
// Vendor detection — id-prefix heuristic + bare-name patterns.
// -----------------------------------------------------------------------------
/** Vendor keys → human label. Add new vendors here when templates pick
* up new model families. */
const VENDOR_LABELS: Record<string, string> = {
"anthropic-oauth": "Claude Code subscription",
anthropic: "Anthropic API",
minimax: "MiniMax",
zai: "Z.ai (GLM)",
moonshot: "Moonshot (Kimi)",
deepseek: "DeepSeek",
"xiaomi-mimo": "Xiaomi MiMo",
openai: "OpenAI",
google: "Google Gemini",
alibaba: "Alibaba Qwen (DashScope)",
nousresearch: "Nous Research (Hermes)",
openrouter: "OpenRouter (any model)",
huggingface: "Hugging Face Inference",
"ai-gateway": "Vercel AI Gateway",
"opencode-zen": "OpenCode Zen",
"opencode-go": "OpenCode Go",
kilocode: "Kilo Code",
"kimi-coding": "Moonshot Kimi (coding-tuned)",
"minimax-cn": "MiniMax China",
"ollama-cloud": "Ollama Cloud",
ollama: "Ollama (self-hosted)",
nvidia: "NVIDIA NIM",
arcee: "Arcee",
xiaomi: "Xiaomi MiMo",
gemini: "Google Gemini",
custom: "Custom OpenAI-compat endpoint",
};
/** Optional per-vendor tooltip shown on hover. */
const VENDOR_TOOLTIPS: Record<string, string> = {
"anthropic-oauth":
"Use your Claude.ai (Pro/Max/Team) subscription via OAuth. Run `claude login` in the workspace terminal to mint the token, then paste it here. No API spend.",
anthropic:
"Pay-per-token via the Anthropic API (Console). Provide an API key starting with sk-ant-…",
minimax:
"MiniMax models served through their Anthropic-API-compatible endpoint. Get a key at platform.minimax.io.",
zai:
"Zhipu AI / z.ai GLM models through the Anthropic-compatible gateway. Get a key at docs.z.ai.",
moonshot:
"Moonshot Kimi K2-series via Anthropic-API-compatible endpoint. Get a key at platform.kimi.ai.",
deepseek:
"DeepSeek V4 via Anthropic-API-compatible endpoint. Get a key at api-docs.deepseek.com.",
openrouter:
"OpenRouter routes to 200+ models behind one API. Use any openrouter/<model> id. Get a key at openrouter.ai.",
huggingface:
"Any model hosted on Hugging Face Inference. Type the full model id (e.g. mistralai/Mistral-7B-Instruct-v0.3).",
custom:
"Self-hosted OpenAI-compatible endpoint (LM Studio, Ollama local, vLLM, llama.cpp). Configure base_url in the workspace's runtime config. No API key required.",
};
/** Sentinel value used in the model <select> for the free-text escape hatch
* added by `allowCustomModelEscape`. The component swaps to a text input
* when this is selected. */
const CUSTOM_MODEL_SENTINEL = "__custom__";
/** Bare-id vendor patterns (no slash separator). Order matters — first
* match wins. */
const BARE_VENDOR_PATTERNS: Array<{ test: (id: string) => boolean; vendor: string }> = [
{ test: (id) => /^minimax-/i.test(id) || /^MiniMax-/.test(id), vendor: "minimax" },
{ test: (id) => /^GLM-/i.test(id), vendor: "zai" },
{ test: (id) => /^kimi-/i.test(id), vendor: "moonshot" },
{ test: (id) => /^deepseek-/i.test(id), vendor: "deepseek" },
{ test: (id) => /^mimo-/i.test(id), vendor: "xiaomi-mimo" },
{ test: (id) => /^claude-/i.test(id), vendor: "anthropic" },
{ test: (id) => /^gpt-/i.test(id), vendor: "openai" },
{ test: (id) => /^gemini-/i.test(id), vendor: "google" },
{ test: (id) => /^qwen-/i.test(id), vendor: "alibaba" },
// Claude-Code OAuth aliases — bare "sonnet"/"opus"/"haiku" + CLAUDE_CODE_OAUTH_TOKEN
// is the strongest signal that this is a subscription model. We also
// gate on env in inferVendor() below to avoid mis-tagging non-OAuth
// models that happen to be named "sonnet".
{ test: (id) => /^(sonnet|opus|haiku)$/i.test(id), vendor: "anthropic-oauth" },
];
/** Infer a vendor key from a model spec. Combines id-prefix and env
* signals. Exported for tests. */
export function inferVendor(model: SelectorModel): string {
const id = model.id || "";
const envSet = new Set(model.required_env ?? []);
// 1. Explicit slash-separated prefix wins (e.g. nousresearch/hermes-4-70b).
const slashIdx = id.indexOf("/");
if (slashIdx > 0) {
return id.slice(0, slashIdx).toLowerCase();
}
// 2. Bare-id pattern. Special-case the OAuth aliases — they only count
// when the env actually demands the OAuth token. Otherwise (e.g.
// a hypothetical "sonnet" alias against ANTHROPIC_API_KEY) fall
// through and let the env-based fallback bucket it under
// "anthropic".
for (const p of BARE_VENDOR_PATTERNS) {
if (!p.test(id)) continue;
if (p.vendor === "anthropic-oauth" && !envSet.has("CLAUDE_CODE_OAUTH_TOKEN")) {
continue;
}
return p.vendor;
}
// 3. Env-tuple fallback. Pick the first env's "namespace" as the
// vendor — e.g. OPENROUTER_API_KEY → "openrouter".
const env = model.required_env?.[0];
if (env) {
const ns = env.replace(/_API_KEY$|_TOKEN$|_KEY$/i, "").toLowerCase();
return ns || "unknown";
}
return "unknown";
}
/** Build the provider catalog from the template's models[]. Models are
* bucketed by `(vendor, sortedEnv)` so two distinct env-tuples for the
* same vendor (rare but possible) become two separate entries. */
export function buildProviderCatalog(models: SelectorModel[]): ProviderEntry[] {
const buckets = new Map<string, ProviderEntry>();
for (const m of models) {
const envs = m.required_env ?? [];
const sortedEnv = [...envs].sort().join("|");
const vendor = inferVendor(m);
const id = `${vendor}|${sortedEnv}`;
const wildcard = m.id.includes("*");
let entry = buckets.get(id);
if (!entry) {
const baseLabel = VENDOR_LABELS[vendor] ?? vendor;
entry = {
id,
vendor,
label: baseLabel,
envVars: envs,
models: [],
wildcard,
tooltip: VENDOR_TOOLTIPS[vendor],
};
buckets.set(id, entry);
}
entry.models.push(m);
// Wildcard sticks if any model in the bucket is a wildcard — same
// bucket can't mix wildcard and concrete because they'd typically
// share required_env but rarely the same vendor. Defensive OR.
entry.wildcard = entry.wildcard || wildcard;
}
// Decorate label with model-count when ≥2 concrete models share the
// bucket. Helps the user understand "Anthropic API (5 models)" vs
// "MiniMax (3 models)".
for (const e of buckets.values()) {
if (!e.wildcard && e.models.length > 1) {
e.label = `${e.label} (${e.models.length} models)`;
}
}
return Array.from(buckets.values());
}
/** Find the provider entry that contains a given model id. Used by
* callers to back-derive the provider when only the model is known
* (e.g. ConfigTab loading from saved state). */
export function findProviderForModel(
catalog: ProviderEntry[],
modelId: string,
): ProviderEntry | null {
if (!modelId) return null;
for (const p of catalog) {
if (p.models.some((m) => m.id === modelId)) return p;
// Wildcard match — entry has model id ending in "*" and the typed
// id starts with the wildcard's prefix (e.g. "openrouter/anthropic/
// claude-3.5-sonnet" matches the "openrouter/*" bucket).
if (p.wildcard) {
for (const m of p.models) {
if (!m.id.endsWith("*")) continue;
const prefix = m.id.slice(0, -1);
if (modelId.startsWith(prefix)) return p;
}
}
}
return null;
}
// -----------------------------------------------------------------------------
// Component
// -----------------------------------------------------------------------------
export function ProviderModelSelector({
models,
value,
onChange,
variant = "stack",
allowCustomModelEscape = false,
disabled = false,
idPrefix,
}: Props) {
const generatedId = useId();
const baseId = idPrefix ?? generatedId;
const providerSelectId = `${baseId}-provider`;
const modelSelectId = `${baseId}-model`;
const catalog = useMemo(() => buildProviderCatalog(models), [models]);
const selected = useMemo(
() => catalog.find((p) => p.id === value.providerId) ?? null,
[catalog, value.providerId],
);
// True when the user picked the "Custom (type model id)..." escape entry
// in the model dropdown — switches to free-text. Wildcard providers
// ALWAYS use free-text, so this flag is for the escape hatch on
// non-wildcard providers.
const userPickedCustom = value.model === CUSTOM_MODEL_SENTINEL || (
!!selected &&
!selected.wildcard &&
!!value.model &&
!selected.models.some((m) => m.id === value.model)
);
const useTextInput = (selected?.wildcard ?? false) || userPickedCustom;
const handleProviderChange = (nextProviderId: string) => {
const next = catalog.find((p) => p.id === nextProviderId) ?? null;
if (!next) {
onChange({ providerId: "", model: "", envVars: [] });
return;
}
// When switching providers:
// - wildcard provider → empty (free-text input takes over)
// - exactly 1 concrete model → auto-pick (no choice to make)
// - 2+ concrete models → leave empty so the operator MUST pick
//
// Background: previously this defaulted to `next.models[0]` for any
// non-wildcard provider, which silently set the alphabetically-first
// model in the bucket. Bit a real user on 2026-05-03 — they picked
// the MiniMax provider intending `MiniMax-M2.7` but the form silently
// set `MiniMax-M2` (first in the list). They never saw the model
// dropdown change because the provider+model widgets are visually
// distinct, and the workspace deployed with the wrong model. Caller
// already disables Deploy/Save while `model.trim() === ""`, so the
// empty default forces an explicit pick without loosening any other
// gate.
const defaultModel = next.wildcard
? ""
: next.models.length === 1
? next.models[0]?.id ?? ""
: "";
onChange({
providerId: next.id,
model: defaultModel,
envVars: next.envVars,
});
};
const handleModelChange = (nextModel: string) => {
if (!selected) {
onChange({ ...value, model: nextModel });
return;
}
onChange({
providerId: selected.id,
model: nextModel,
envVars: selected.envVars,
});
};
const containerClass = variant === "grid" ? "grid grid-cols-2 gap-3" : "space-y-3";
return (
<div className={containerClass} data-testid="provider-model-selector">
<div>
<label
htmlFor={providerSelectId}
className="text-[10px] uppercase tracking-wide text-ink-mid font-semibold mb-1.5 block"
>
Provider <span aria-hidden="true" className="text-bad">*</span>
<span className="sr-only"> (required)</span>
</label>
<select
id={providerSelectId}
value={value.providerId}
onChange={(e) => handleProviderChange(e.target.value)}
disabled={disabled || catalog.length === 0}
aria-describedby={selected?.tooltip ? `${providerSelectId}-help` : undefined}
data-testid="provider-select"
className="w-full bg-surface-sunken border border-line rounded px-2 py-1.5 text-[11px] text-ink focus:outline-none focus:border-accent focus:ring-1 focus:ring-accent/20 transition-colors disabled:opacity-50"
>
<option value="" disabled>
select provider
</option>
{catalog.map((p) => (
<option key={p.id} value={p.id} title={p.tooltip}>
{p.label}
</option>
))}
</select>
{selected?.tooltip && (
<p
id={`${providerSelectId}-help`}
className="text-[9px] text-ink-mid mt-1 leading-relaxed"
>
{selected.tooltip}
</p>
)}
{selected && selected.envVars.length > 0 && (
<p className="text-[9px] text-ink-mid mt-0.5 font-mono">
requires: {selected.envVars.join(", ")}
</p>
)}
</div>
<div>
<label
htmlFor={modelSelectId}
className="text-[10px] uppercase tracking-wide text-ink-mid font-semibold mb-1.5 block"
>
Model <span aria-hidden="true" className="text-bad">*</span>
<span className="sr-only"> (required)</span>
</label>
{useTextInput ? (
<>
<input
id={modelSelectId}
type="text"
value={
value.model === CUSTOM_MODEL_SENTINEL ? "" : value.model
}
onChange={(e) => handleModelChange(e.target.value.trim())}
placeholder={
selected?.wildcard
? wildcardPlaceholder(selected)
: "type any model id"
}
disabled={disabled || !selected}
spellCheck={false}
autoComplete="off"
data-testid="model-input"
className="w-full bg-surface-sunken border border-line rounded px-2 py-1.5 text-[11px] text-ink font-mono focus:outline-none focus:border-accent focus:ring-1 focus:ring-accent/20 transition-colors disabled:opacity-50"
/>
<p className="text-[9px] text-ink-mid mt-1 leading-relaxed">
{selected?.wildcard
? wildcardHelpText(selected)
: "Free-text model id. Make sure the provider can resolve it."}
</p>
{!selected?.wildcard && (
<button
type="button"
onClick={() => {
// Switch back to dropdown by setting model to first
// concrete option.
if (selected) {
handleModelChange(selected.models[0]?.id ?? "");
}
}}
className="text-[9px] text-accent hover:text-accent mt-0.5 focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-accent focus-visible:ring-offset-1 focus-visible:ring-offset-surface rounded"
>
back to model list
</button>
)}
</>
) : (
<select
id={modelSelectId}
value={
value.model && selected?.models.some((m) => m.id === value.model)
? value.model
: ""
}
onChange={(e) => {
if (e.target.value === CUSTOM_MODEL_SENTINEL) {
handleModelChange(CUSTOM_MODEL_SENTINEL);
} else {
handleModelChange(e.target.value);
}
}}
disabled={disabled || !selected || selected.models.length === 0}
data-testid="model-select"
className="w-full bg-surface-sunken border border-line rounded px-2 py-1.5 text-[11px] text-ink font-mono focus:outline-none focus:border-accent focus:ring-1 focus:ring-accent/20 transition-colors disabled:opacity-50"
>
<option value="" disabled>
{selected ? "— select model —" : "— select provider first —"}
</option>
{selected?.models
.filter((m) => !m.id.includes("*"))
.map((m) => (
<option
key={m.id}
value={m.id}
title={m.name ?? m.id}
>
{m.name ?? m.id}
</option>
))}
{allowCustomModelEscape && selected && (
<option value={CUSTOM_MODEL_SENTINEL}>
Custom (type model id)
</option>
)}
</select>
)}
</div>
</div>
);
}
function wildcardPlaceholder(p: ProviderEntry): string {
const example = p.models.find((m) => m.id.includes("*"))?.id ?? "";
if (!example) return "type any model id";
// Strip trailing star — show the pattern as a hint.
const prefix = example.replace(/\*$/, "");
switch (p.vendor) {
case "huggingface":
return `e.g. ${prefix}meta-llama/Meta-Llama-3-70B-Instruct`;
case "openrouter":
return `e.g. ${prefix}anthropic/claude-3.5-sonnet`;
case "custom":
return `e.g. ${prefix}my-local-model`;
default:
return `e.g. ${prefix}<model-id>`;
}
}
function wildcardHelpText(p: ProviderEntry): string {
switch (p.vendor) {
case "huggingface":
return "Any model hosted on Hugging Face Inference. Browse at huggingface.co/models?inference=warm.";
case "openrouter":
return "Any of OpenRouter's 200+ routed models. Browse at openrouter.ai/models.";
case "custom":
return "Self-hosted endpoint. Configure base_url in your workspace's runtime config (no API key required).";
case "ai-gateway":
return "Vercel AI Gateway model id. See vercel.com/docs/ai-gateway.";
case "opencode-zen":
return "OpenCode Zen model id. See opencode.zen.";
default:
return "Wildcard provider — type the model id in full. Provider routes by id prefix.";
}
}